A modern programming language that makes developers happier.

kotlinlang

In the last week I've released a couple of updates to Chasm which have over doubled its performance, previously we we're hitting 220 ish on the industry standard coremark benchmark and we're now hitting over 500. Chasm was already the fastest Wasm interpreter on the JVM but now we're really in our own lane. We're fast approaching the ceiling of what can be achieved inside of JVM bytecode. If you're interested in any of the details theres summaries in the release links attached

<https://github.com/CharlieTap/chasm/releases/tag/1.4.0>
<https://github.com/CharlieTap/chasm/releases/tag/1.4.1>

Do you have some numbers how this relates to chasm with AOT compilation running on a JITed vs. a non-JITed VM? I am just curious :-)

<@U01ERMR974M> Off the top of my head, Chasm hits around 520. The fastest wasm interpreter in the world right now hits around 5000, and the AOT wasmtime around 20000.

All the above are relative to my machine (m4 max).

What I find in practice (at least with respect to apps) is that workloads in the single digit millis and lower are suitable but often anything over that is not.

We are fundamentally limited on the jvm by the bytecode available and cannot write what is considered an optimal loop (direct threaded), realistically I can maybe hit 600 ish with in the current constraints. I've been working on a new runtime in a low level language, and by just writing the optimal loop I was able to hit 1500 without any optimisations